DCE-MRI-based radiomics model explained by SHAP method for predicting CDKN2A/B homozygous deletion in IDH-mutant Astrocytomas

Motivation: Cyclin-dependent kinase inhibitor (CDKN) 2A/B homozygous deletion holds an important value in the prognostic analysis of IDH-mutant astrocytomas. Goal(s): This study aimed to develop a radiomics model based on dynamic contrast-enhanced (DCE) MRI parameter maps for predicting CDKN2A/B homozygous deletion status in IDH-mutant astrocytomas. Approach: Radiomic features extracted by Pyradiomics, and selected by independent samples t-test and LASSO algorithm. Prediction models were constructed by support vector machine, and explained by Shapley Additive exPlanations method. Results: The radiomics model based on mixed features from all DCE parameter maps exhibited AUC of 0.957 and 0.924 in the training set and validation set, respectively. Impact: Our study proved that the DCE-MRI-based radiomics model has excellent ability to predict CDKN2A/B homozygous deletion status in IDH-mutant astrocytomas.